Submitted Ideas
0
Automate habitat classification for deepwater monitoring
Posted by Dirk Rosen | , United States
The advent of remotely operated and autonomous underwater vehicles (ROVs and AUVs) has enabled humans to systematically and more cost effectively document deepwater habitats and biotic communities beyond diver depths to detect change over time. However the cost of post processing and analyzing this video and digital stills data could be substantially lowered. Approximately half of a deepsea survey project cost is the post-processing of the video imagery itself. Our organization, Marine Applied Research and Exploration, has recently completed a multi-year effort in partnership with the California Department of Fish and Wildlife to characterize deepwater habitats and ecosystems of California’s network of marine protected areas. Deploying ROVs we have surveyed over 2,300 kilometers of seafloor in 275 sites along California’s entire coast, amassing thousands of hours of video. In order to transform this video footage collected at depth into accurate, geo-referenced and reliable descriptions of deepsea habitats and ecosystems, skilled individuals must scrutinize every second of video footage to identify characterize and enumerate physical features and living organisms. Our team has now post-processed thousands of hours of video and identified (and in many cases sized) over two million individual fish and invertebrates for California Department of Fish and Wildlife, which will enable resource managers to scientifically manage our deep-ocean resources. This post-processing is labor-intensive and somewhat costly given todays’ technologies. This cost remains a barrier to some resource management agencies that might otherwise jump at the opportunity to acquire more accurate information about the marine ecosystems they govern. Numerous attempts to automate (or partially automate) benthic video post-processing at world renown institutions have been largely unsuccessful. We believe that the root cause of these failures has been that researchers have focused their efforts on the most challenging targets – fish and mobile invertebrates – which can be notoriously difficult. A better approach would be to begin with the stationary elements – physical habitat and sessile invertebrates – automatic characterization and identification of which will be comparatively easy to develop. When successful, this would reduce the cost of post-processing by thirty to forty per cent (30%-40%), and allow those savings to be re-purposed into either further survey of the 98% unexplored depths beyond divers, or performing deeper levels of data analysis. Either way this enhances resource managers’ understanding of the opaque, poorly sampled world under their jurisdiction.
